System Programmer Reference › Utilities › VMBMCOPY Utility › Performance Notes

Performance Notes

VMBMCOPY DUMPJOB and TAPESET copy jobs are much slower than when CA VM:Backup produces the original backups. This slowness is due to the following reasons:

• Disk I/O is usually faster than tape I/O. Although this generalization applies more to older technology, it is still true today in some situations. A backup job is at least partly disk I/O and so always runs faster than a tape-to-tape copy.
• VMBMCOPY uses CMS tape I/O facilities, which do diagnose I/O. Diagnose I/O is low on overhead but is inherently single-threaded. In other words, when a tape read or write is invoked through a diagnose instruction, the instruction does not complete until the I/O completes. In contrast, CA VM:Backup performs its own I/O and is able to utilize fully the tape hardware by driving multiple reads and writes simultaneously.
• A CA VM:Backup backup job can be broken into as many as 15 separate streams, which run independently. Using multiple streams improves the utilization of the hardware even further. VMBMCOPY processes input streams serially, producing what is in effect a single output stream. Serial streaming improves the resulting tape utilization but adds greatly to the elapsed time necessary to copy all the tapes.

All of these reasons result in lengthy copy operations. This time cost is the primary reason that we recommend that you use multiple VMBMCOPY user IDs to copy backup tapes.